How are you catching agent tests that pass on the old code?text

A developer is seeking methods to catch agent tests that pass on old code, noting that tools like Cursor and Claude incorrectly mark tests as green. They currently use a GitHub Action for pytest and found mutation testing on every PR to be too slow. The developer is asking how others are addressing this issue, specifically if they are gating tests or just reviewing test files.

This post highlights a specific challenge with AI-assisted development tools like Cursor and Claude, which unlike traditional testing methods, can incorrectly mark tests as passing.